Immigration to Norway 1969-2010 : effects of policies and EEA membership

Discussion Papers

Statistics Norway Research department No. 687

•

March 2012

Ådne Cappelen and Terje Skjerpen

Immigration to Norway 1969-2010

Effects of policies and EEA membership

Discussion Papers No. 687, March 2012 Statistics Norway, Research Department

Ådne Cappelen and Terje Skjerpen

Immigration to Norway 1969-2010 Effects of policies and EEA membership

Abstract:

We examine how changes to regulations and the economic conditions have influenced gross immigration to Norway from, in principle, all countries in the world during 1969– 2010. In line with existing studies of immigration we find that economic factors were important for immigration to Norway. Income differences between Norway and other countries have the expected impact, as do changes in income distributions. The labour market situation has also been important in that lower unemployment in Norway has resulted in higher immigration and higher unemployment in the country of origin has led to higher emigration to Norway. We find that immigration policies have largely had the expected effects. One example is the 1975 ‘immigration halt’ that did have a strong and long lasting effect on total immigration to Norway. Further tightening of the immigration regulations that came in 1977 also reduced immigration, while the more liberal policies introduced in 1981

contributed to higher immigration. From 2000 to 2010 several changes linked to the enlargement of EU influenced immigration to Norway. Norway’s membership in the European Economic Area (EEA) in 1994, and in the Schengen-area in 2001 resulted in higher immigration while the 2004 and 2007 EU enlargements also increased labour immigration to Norway substantially.

Keywords: Immigration, Immigration policies, Incentive variables JEL classification: J11, J15

Acknowledgements: We would like to thank Norwegian Directorate of Immigration (UDI) for financial support, Jørgen Ouren for help with the data and Taryn A. Galloway, Eivind Hoffmann (at the UDI), Arvid Raknerud and Lars Østby for useful comments.

Address: Terje Skjerpen, Statistics Norway, Research Department, P.O. Box 8131, Dep. N-0033 Oslo Norway. E-mail: terje.skjerpen@ssb.no

Ådne Cappelen, Statistics Norway, Research Department, P.O. Box 8131, Dep. N-0033 Oslo Norway. E-mail: cap@ssb.no

Discussion Papers comprise research papers intended for international journals or books. A preprint of a Discussion Paper may be longer and more elaborate than a standard journal article, as it may include intermediate calculations and background material etc.

© Statistics Norway

Abstracts with downloadable Discussion Papers in PDF are available on the Internet:

http://www.ssb.no

http://ideas.repec.org/s/ssb/dispap.html

For printed Discussion Papers contact:

Statistics Norway

Telephone: +47 62 88 55 00 E-mail: Salg-abonnement@ssb.no

ISSN 0809-733X Print: Statistics Norway

Sammendrag

Innvandringen til Norge har økt gradvis og netto innvandringen har gjennomgående vært positiv og økende siden 1970. Etter utvidelsen av EU i 2004 har innvandringen skutt fart og har svingt rundt 70 000 i senere år. På 1970-tallet iverksatte myndighetene tiltak for å dempe innvandringen til Norge.

På 1980-tallet var det en viss liberalisering igjen av politikken. Siden 1994 har endringen i innvand- ringen blitt sterkt påvirket av Norges tilknytning til EU. I denne studien analyserer vi hvordan ulike politiske tiltak og endringer i økonomiske omstendigheter har påvirket innvandringen til Norge. Tall for brutto innvandring fra i prinsippet alle land i verden til Norge fra 1969 til 2010 studeres.

I økonomisk forskning om migrasjonsstrømmer finnes det en standardmodell for individers flyttebeslutninger. Modellen vektlegger økonomiske forhold i hjemlandet sammenliknet med forholdene dit man vurderer å flytte. Forskjeller i hva man vil tjene spiller en rolle, men også mulighetene for å få seg arbeid dit man kommer betyr noe. Kostnadene ved å flytte og etablere seg spiller åpenbart en rolle for om det er verd å flytte. Her kommer kulturelle og språklige forskjeller inn.

I noen sammenhenger har økonomiske forhold liten betydning for beslutningene fordi man flykter av politiske grunner fra ett land til andre land, eller det kan være familiære bånd som motiverer flytting.

I tråd med mange studier av innvandring finner vi at økonomiske bakgrunnsvariabler har betydning for innvandring til Norge. Inntektsforskjellene mellom Norge og utlandet inngår med det forventede fortegnet og også forskjeller i fordelingen av inntekt spiller en rolle. Jo skjevere inntektsfordelingen i Norge er sammenliknet med i opprinnelseslandet, jo større blir innvandringen. Også arbeidsmarkeds- situasjonen i Norge har betydning. Er arbeidsløsheten i Norge lav, vil det komme flere til Norge. Vi har kun data for arbeidsmarkedssituasjonen i noen av landene vi studerer, men for disse viser resultatene at høyere ledighet i opprinnelseslandet øker innvandringen til Norge.

Vi finner også at mange innvandringspolitiske tiltak har hatt den forventede effekten. Det gjelder for eksempel innvandringsstoppen som formelt ble innført i 1975. Vi har estimert at dette inngrepet hadde en stor og langvarig betydning for samlet innvandring til Norge. Også den videre innstramming i regelverket som skjedde i 1977 har dempet innvandringen, mens liberaliseringen i 1981, som

forventet, bidro til høyere innvandring enn hva vi ellers ville ha fått. I tiden rundt 1990 var det mange spesielle begivenheter som påvirker innvandringen til Norge. Vi finner at både norsk deltakelse i EØS i 1994 og Schengen-avtalen fra 2001 bidro til økt innvandring, men særlig utvidelsen av EU i 2004 har hatt stor betydning for innvandringen til Norge. EU-utvidelsen i 2007 har ytterligere økt

innvandringen til Norge. Innstramming i reglene for familiegjenforening i 2008 har hatt en betydelig effekt på innvandringen ifølge vår analyse.

1. Introduction

Immigration to more developed nations has increased significantly for several decades. In Europe the breakdown of the “iron curtain” affected migration flows, as has EU enlargements, which brought many former East-European countries into a common labour market. Although Norway is not a member of the EU, it is part of the European Economic Area (EEA) and consequently part of the common European labour market. Norway is thus affected by migration flows in Europe just as any other EU-country and migration to Norway from EU-countries has increased significantly in recent years. While Norway historically was a country with more emigration than immigration, the opposite has been the case more recently. Indeed, Norway together with Ireland was one of the countries with the highest rate of emigration during last decades of the 19^th century and the first decade of the 20^th century. This changed with more restrictive immigration policies in the US from the 1920s and the depression of the 1930s. Until around 1970 net immigration to Norway was negative or small. From around 1970 net immigration has been positive and gradually increasing, cf. Figure 1. With a total population of roughly 4.9 million in 2010 net immigration increased the Norwegian population by 0.8 percent that year.

Figure 1. Migration to Norway. 1951−2010

- 10 000 0 10 000 20 000 30 000 40 000 50 000 60 000 70 000 80 000

1951 1955 1959 1963 1967 1971 1975 1979 1983 1987 1991 1995 1999 2003 2007 Immigration

Emigration Net immigration

Figure 2 shows the reported motives for immigration to Norway since 1990, when the collection of these statistics started, as defined by reason for the residence permit granted.¹ We see that the number of persons admitted following an application for asylum has varied around a fairly constant level. The

1 From 1 October 2009 non-Nordic citizens of the European Union (except Bulgarians and Romanians) only need to declare the main purpose of the stay when registering with the Norwegian authorities.

peak in 1993 is related to the war in Bosnia while the 1999-peak is mostly related to the Kosovo conflict. Student immigration to Norway has been steadily increasing from a low level. The number of people who come for work used to be at the same, quite low level, but has increased dramatically since the expansion of the EU in May 2004. Family reunion has been an important reason for immigration but is related to the other reasons and in particular to those who come for work or seeking protection.

Note that these statistics do not include immigrants who are citizens of another Nordic country because they have had free access to Norway since 1957 and do not have to state any reason for immigrating when registering with the Population register.² Also, the statistics do not include intended stays of less than six months.

Figure 2. Immigration to Norway by registered reason for immigration¹ 1990-2009

0 5 000 10 000 15 000 20 000 25 000

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 Work

Family Refugee Education

1 Does not include citizens of the other Nordic countries (Denmark, Finland, Iceland and Sweden)

From the mid 1970s immigration policy became a new theme in Norwegian politics and attempts at restricting immigration were put in place by a new law. Later various additional measures have been introduced, but not all of these have been restrictive. Some have been of a more liberal nature. From 1994 and onwards immigration into Norway has been affected by Norway joining the EEA. In this paper we study the effects of various immigration policy measures on immigration to Norway from all countries in the world using a macro data panel from 1969 to 2010. Policies have not been uniform across countries so we specify and test country specific or region specific policies. In order to do this we translate various immigration policies into a set of dummies for each policy. We shall return to this in Section 3 of the paper.

2 Citizens of other countries need a residence or work permit and the basis for granting the permit is registered by UDI.

There are many studies that analyse migration based on a single destination country. The study by Clark et al. (2007) for the United States and by Hatton (2005) for the United Kingdom, both find evidence for the role of immigration policies. Karemera et al. (2000) study migration to North

American destinations while Mayda (2010) studies migration to 14 OECD countries. See also Massey et al. (1993) for a description of various theories of migration. A number of variables have been suggested as driving forces in these migration studies. Some relate to cultural and linguistic factors while other take on a more economic perspective and focus on differences in economic opportunities such as income and labour market features. Our main focus is to analyse how changes to Norwegian immigration policies have influenced migration to Norway during the previous four decades. We incorporate some of the main ideas in previous studies of migration, and test if migration policies in Norway can explain some of the changes in immigration flows over time and from particular countries or groups of countries. Using a panel of 179 countries with statistics from 1969 to 2010 we conclude that not only do economic variables explain changes in migration to Norway over time but some of the major policy changes that have taken place are also important in understanding immigration to

Norway.

In the next section we present our modelling framework while the third section discusses the data and in particular how we have created the policy intervention dummies that are linked to various

immigration policies. The fourth section presents our main results and a number of sensitivity tests.

We conclude in section five.

2. Modelling framework

Our basic model dates back to Roy (1951) and is elaborated by Borjas (1987). For a recent application see Mayda (2010). There are two countries: (o)rigin and (d)estination. The log of wages that an individual living in the origin country would receive if not migrating (wo) is assumed to be

log(w_o)=

μ ε

_o+ _o, ^where ε_o ^~N(0,σ_o^2). (1) Here

μ

₀ is interpreted as determined by individual observables such as education, gender etc., while

ε

0 captures unobservable characteristics with zero mean and a constant variance. For individuals who immigrate there is a similar wage model in the destination country

log(w_d)=

μ

_d +

ε

_d, ^where

ε

_d ^~N(0,

σ

_d^2). (2)

The error terms are possibly correlated with a correlation coefficient

ρ

. Hatton (2005) and Clark et al. (2007) let the

μ

'sdepend linearly on skill which is also assumed to be distributed normally so that

log(w_o)and log(w_d)are also normally distributed.

The decision to immigrate or not is determined by the sign of an index I :

( ) ( )

log _d ( _o ) _{d o d o}.

I = w w +c ≈

μ

−

μ δ

− +

ε

−

ε

(3)

Here c is the level of mobility costs while δ is the wage equivalent mobility cost. Immigration occurs if the value of the index I is positive. Summing over all individuals in the origin country, the emigration probability, P, from the origin country is given by

( )

( ) ( ^{( )} ) ( ^{( )} )

Pr _{d o d o} 1 _{d o} / _{d o} / .

P=

ε

− > −

ε μ

−

μ δ

− = −

Φ

−

μ

+

μ δ σ

+ _ε =

Φ μ

−

μ δ σ

− _ε (4)

Here,

σ

_ε is the standard deviation of the difference of the error terms,

ε ε

= _d −

ε

_o,^and

Φ

^{is the}

standard normal cumulative distribution function. Equation (4) captures some important features of empirical models of immigration. Higher income in the origin country lowersP, while higher income in the destination country increases P. In addition, the income effects are the same but with opposite signs. The variance of εis given by

2 2 2

2 .

d o do

σε =σ +σ − σ (5)

If the destination country has a more equal distribution of income than the origin country, and this would usually be the case when Norway is the destination country, an increase in inequality in the destination country will lower

σ

_ε ^.3 If the term in the brackets in (4) is negative so that the income in the destination country is higher than in the origin country adjusted for migration costs, an increase in destination inequality will increase immigration as argued for by Borjas (1987), Hatton (2005), and Clark et al. (2007). Borjas (1987) was the first to include the income distribution as a feature affecting migration. He finds that countries with more income inequality have lower emigration rates. For this

3 Note that ∂σε ∂σ^d =

(

σ^d −σ σ^o

)

ε ^when

ε

_d ^and

ε

o are assumed to be perfectly correlated.

to be the case there must be a strong positive correlation between earnings for immigrants in the origin and the destination countries and less income inequality in the destination country. If the mean income in the destination country is higher than in the origin country – which is a major motive for emigration in the first place – and inequality increases in the origin country, then high-income persons in that country will have fewer incentives to emigrate while low-income persons in the origin country are not affected. Total emigration is then reduced. Thus, changes in the distribution of income in the origin country select or motivate on average different people to emigrate.⁴ Mayda (2010) argues for including also a quadratic term of relative income inequality and finds empirical support for this specification.

Also Hatton (2005) and Clark et al. (2007) find significant effects of variables characterising the income distribution in their models.

Pin (4) is the emigration probability defined as emigration divided by the relevant population in the origin country or the emigration rate. If a model is specified using the number of emigrants as the endogenous variable while the size of the population of the origin country enters as a regressor, this restriction can be tested. This is done by Karemera et al. (2000) who include the (log) population in the emigration equation but their results do not support using the emigration rate specification. Kim and Cohen (2010) combine the specification in (4) into a gravity model. Let M_oddenote the number of immigrants at any time from country oto country d, P_o the population of the origin and P_dthe population in the destination, the simplest gravity model is

, ,

od o d od

M =kP P d o≠d (6)

wherekis a constant andd_odrefers to the distance between oandd. The standard specification used is achieved by dividing byP_oon both sides of Eq. (6) so the added feature of the gravity model is really the inclusion of the population of the destination country. Kim and Cohen (2010) test the restriction of unit elasticities of the population terms in the equation and generally reject the restriction; although in several versions their estimate of the elasticity of P_o is not far from one.⁵

4 When

σ

_ε goes towards infinity it follows from (4) that the emigration probability goes towards 0.5. Thus in this case the individual acts as if he tosses a coin whether he should emigrate or stay.

5 However, in the current paper we do not relate our model to the gravity specification.

Higher monetary costs of migration relative to income in the destination country reduce migration according to the model in (4). A theoretical model of the effects of mobility costs is the focus of Carrington et al. (1996). The idea here is that mobility costs decrease with the number of migrants already settled in the destination country because they send information about job and housing markets to friends and family in the origin country and generally provide a network for new entrants. The empirical specification of mobility costs is a central part of econometric analyses of immigration.

Standard proxies used are language differences, geographical distance, and migration policy indicators. It is common to include social indicators like crime and corruption indicators of political systems in order to explain migration flows. Several studies referred to earlier use more or less these variables in their econometric specifications. We proxy these factors using the number of resident immigrants by country divided by the Norwegian population as one indicator for migration costs. In addition our model includes fixed effects for all countries to capture other country specific factors. We also allow for these factors to change over time by including country specific time trends.

3. Data and specification of immigration policies

Statistics on demographic and economic variables

Statistics for immigration to Norway from every country in the world are readily available in the statistics database, “Statbank” on the webpage of Statistics Norway.⁶ We have chosen to model immigration by country of departure and not citizenship. Statistics on immigrants by citizenship are available, but that series starts much later and makes the study of immigration policies before 1986 impossible.⁷ Also it is not entirely clear what to prefer in our context. An Ethiopian who has lived in Sweden for some time may just as well be motivated by the same factors as a Swede even if the policies that apply to him/her are different (if (s)he has not acquired Swedish citizenship). Statistics on the stock of immigrants by country is also found in this database.

For a number of the countries in the world, immigration to Norway does not take place every year. In fact for some small islands in the Pacific and Caribbean migration to Norway is a rare event. To take one example: during the period 1969 – 2010 there are four years of recorded immigration to Norway from Samoa. In these cases we have simply excluded the country from our list. We have also excluded countries where immigration never reaches 5 persons in any year. For some countries where

6 http://www.ssb.no/english/subjects/00/00/10/innvandring_en/

7 With one exception noted below the fact that the regulations apply to country of citizenship and not of previous residence is not expected to influence the results.

immigration is quite regular, there are also some years with no recorded immigration. These zero observations have been excluded from the sample in line with Kim and Cohen (2010). Table B2 shows the number of observations by country included in the sample.⁸

Population statistics for all countries have been taken from United Nations, Population Division.⁹ The statistics for Norway have been taken from the Statbank, as referred to earlier.

For economic statistics we rely on relative income measured by GDP per capita in PPPs and current US dollars based on Penn World Tables cf. Heston et al. (2011). We use GDP-figures in nominal terms as it is relative GDP-levels that are used in the model. We have also included the unemployment rate in Norway.¹⁰ These figures are taken from OECD-databases and usually go back to 1970. Income data are problematic. We have relied on three main sources of information. High quality data,

sometimes even going back before 1970, are generally available for countries taking part in the Luxembourg Income Study (LIS).¹¹ For most countries however, we rely on the WIDER database.¹² For Latin-American countries we also use data from the SEDLAC homepage.¹³ The WIDER database indicates data quality by using a scale from 1 to 4. When possible we rely mostly on high quality data but have tried to make our coverage as complete as possible. In general, data are better and

comparable the more recent they are. For some countries there are comparable figures only for a few years. These are used to calibrate the level and lower quality data are used to interpolate between these years. When also these are missing linear interpolation is used.

Immigration policies and legislation in Norway

We now turn to how we have translated Norwegian immigration policies into quantitative variables.

First, we emphasise that immigration from the other Nordic countries (Denmark, Finland, Iceland and Sweden) has not been affected by any policy changes after the establishment of a Nordic passport union in 1957, which gave Nordic citizens free access to all the Nordic countries without needing

8 In Table B4 we list the countries that are excluded from our analysis.

9World Population Prospects: The 2008 Revision - Extended CD-ROM Edition.

WPP2008_ASCII_FILES/WPP2008_DB02_POPULATIONS_ANNUAL

10In some subsample estimations we also exploit the unemployment level in the origin country. For many countries in the sample no reliable unemployment data have been found and the sample where unemployment in the origin country is included is therefore much smaller than the total sample.

11 Data can be found on http://www.lisdatacenter.org/data-access/keyfigures/

12 Cf. UNU-WIDER World Income Inequality Database, Version 2.0c, May 2008 available at http://www.wider.unu.edu/.

LIS data is also included in the WIDER database.

13 http://sedlac.econo.unlp.edu.ar. Database updated by April 2011.

passports, resident permits or work permits. It is also possible for Nordic citizens to commute or immigrate to Norway for short term stays, e.g. to work, without even having to register with the population register that represents the main source of the immigration statistics used in this study.

Consequently, no changes in immigration policies affect Nordic citizens.

Table 1. An overview of policy dummies and their expected sign in the econometric model DDUM1974 Ban on general work permits. All countries. Negative

DUM1977 Residence permits not granted to illegally entrants. All countries. Negative DUM1981 Residence permits for immigrant students and school attendants. They were also

given work permits. More liberal rules for family reunions. All countries. Positive DUM1991 Easier family reunion, work permits given to applicants for residence. All countries.

Positive

DUM1994 Norway joins the EEA. EEA-citizens free access. Positive

DUM1997 Liberalisation related to the Geneva-convention. Refugees. Positive DUM1998 Liberalisation for refugees. Positive

DUM1999 New law on human rights. UN convention on women and children. Positive DUM2000A Easier access for people with specialist competence. Positive

DUM2000B Easier access for Iraqis. Positive

DUM2001A Schengen-convention. Liberalisation for Schengen member countries (”S”) DUM2001B Schengen-convention may affect immigration from non-Schengen countries (“O”)

negatively

DDUM2003 Liberalisation in 1997 tightened in 2003. Affecting mostly people from Afghanistan, Iraq, Somalia and countries in former Yugoslavia. Negative

DDUM2004 Extension of EU included Czech republic, Cyprus, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovakia and Slovenia. Positive for these countries DUM2005 Easier access for Vietnamese refugees on the Philippines and Iraqis. Positive DUM2006 More restrictive rules for family reunion for immigrants arriving on tourist visa.

Negative for non-EU countries

DUM2007EU New EU members: Bulgaria and Romania. Positive for these two countries DUM2007A New EU members from 2004 included in the Schengen area. Positive DUM2007B Residence for certain asylum seekers. Positive

DUM2008 Stricter economic demands for family reunion. Negative

DUM2009A Temporary and transition rules applying to new (from 2004) EU members lifted.

Positive effect for countries affected by Dummy 2004.

DUM2009B Switzerland joins Schengen. Positive

Out of a large number of changes to laws and regulations listed on the home page of the Norwegian Directorate of Immigration¹⁴ we have selected 21 as basis for specifying policy dummies to capture various aspects of policy changes, where some changes apply to all countries, some to a group of countries and, sometimes only to very few or even a single country. Since some of the policy changes are partly overlapping in time, one cannot include too many of the policy dummies in the model specification. Table 1 summarises the policy variables included in our study. We have included what we regard as the most important policy changes. We disregard minor changes such as higher visa fees

14http://www.udi.no/Oversiktsider/Statistikk-og-analyse/FoU---rapporter1/Historisk-oversikt-over-regelverksendringer-/.

(which are anyway quite moderate). A certain element of subjectivity must of course be used when choosing what to include and what to exclude and here we have relied on expert advice from the immigration authorities in our selection of policy changes. The presentation of the policy changes in Table 1 gives an idea of the level of detail that we address and implicitly what we have excluded in the sense that other changes are not judged as being important enough on a priori grounds relative to those we have included. We refer to Appendix C for a more detailed explanation of the policy dummies. We should also note by specifying changes as step dummies we cannot be sure that we actually capture a policy change. The step dummies could in principle capture other changes affecting immigration. We do try to address this issue to some extent with robustness checks in Section 4.

4. Model and empirical results

In this paper we consider a specification of the following type:¹⁵

. I i , 2010 ,..., 1969 t

; ) 1966 t

( DUMSOMALIA

DUMLIBERIA

DUMCHILE INTERV

GINIRATIO GINIRATIO

UR URNOR

) /GDPCAPNOR log(GDPCAP

) /PNOR log(IS

) P / M log(

) P / M log(

145

it i

i t 3

t 2

t 1

it 2

it 7

it 6

1 - t i, 5 1 - t 4

2 - t 2

- t i, 5

1 - t 1

- t i, 2 1 t, i 1 t, i 1

t, i t, i

∈

=

+

− + + +

+ +

+ +

+ +

+

+ +

= _{− −}

ε δ

μ ρ

ρ

ρ γ β

β

β β

β

β β

(7)

The left hand side variable in Eq. (7) is the log of the (scaled) gross immigration rate (gross immigration to Norway divided by the population) of country i in year t. As explanatory variables we have, in addition to the lagged endogenous variable, six “incentive” variables and a country-specific vector of intervention variables, INTERV_it. Finally we have added (i) some dummy variables to account for large residuals for some countries (Chile, Liberia and Somalia), (ii) country specific fixed effects,

μ

,(iii) linear country- specific deterministic trends,δ , and (iv) a genuine white noise error term, ε. The incentive variables are:

(i) the log of the ratio between the immigration stock of country i and the Norwegian population (IS/PNOR) lagged one year, to capture effects on immigration costs in that a higher number of previous immigrants from a country will make it less costly for newcomers to settle in Norway, cf. Carrington et al. (2003),

15 We specify (7) with lags on the incentive variables corresponding to what we have ended up with in connection with our reference model. The lags were chosen such that the incentive variables entered the equation with the correct sign and as significant as possible. Furthermore, one may argue, from a theoretical point of view, that the immigrants probably need some time to assess relevant information in conjunction with an immigration decision.

(ii) the log of GDP per capita of country i divided by GDP per capita for Norway lagged two years, (GDPCAP/GDPCAPNOR) in order to capture the relative income effect,

(iii) the unemployment rate in Norway (URNOR) lagged one year, to capture the effect of labour market slackness on migration,

(iv) the unemployment rate in the origin country to capture the effects of labour market condition as a push factor behind emigration,

(v) a second order polynomial in the ratio between the income distribution in the origin country and the income distribution of Norway to capture a selection mechanism behind immigration.

The variables contained in the vector with intervention variables,INTERV, are those occurring in the text column of Table 2 after the variable DUMSOMALIA. For further information on these variables cf.

Table 1, Table B1 and Appendix C. It is not uncommon to include fixed period effects in panel data models. However, in the current we have decided not to do so. Since the models already contain various time dummies that are assumed to influence the vast majority of countries and since all the model specifications involve the Norwegian unemployment rate, inclusion of year dummies will raise questions of identification and lead to overparameterized models with unclear interpretation.

I145denotes a set with 145 current country numbers that are listed in Table B2. The panel data set is unbalanced and Table B2 gives an overview of the effective number of observations for each country in

I145. We have, as noted earlier, omitted some small countries and observations for which the number of immigrants to Norway in the current and previous year is less than five persons.¹⁶

Main Empirical results

Since we have information on foreign unemployment and income distribution only for a subset of countries these variables are not included in our basic model.¹⁷ Weighted least squares, with weights based on population size, is our main estimation method, but we also present estimates based on ordi- nary least squares.¹⁸ We have not tested the assumption of exogenous policy dummies. Heuristically

16 Cappelen et al. (2011), who use the same data material as in the current paper, consider subsample estimation for different geographical areas. Such type of robustness analysis is not undertaken in this paper, but is viewed as a topic for further work.

17 The reference model corresponds to the restricted case in Table 2.

18 All the calculations have been done by means of TSP version 5.0, cf. Hall and Cummins (1995). This software program contains a module for panel data analysis. However, this routine has not been utilized since we (i) consider weighted regression and (ii) incorporate country-specific linear deterministic trend effects. Thus, we have estimated the model using the routine for weighted least squares. This is facilitated by including a large amount of deterministic variables that take care of country specific effects and country specific linear trends. We do not consider random effects models in this paper. Consistent estimation of random effects models with lagged endogenous variables requires instrumental variables. We leave this for future analysis.

we do have some arguments in support of exogeneity. All variables in the main model except the un- employment and the policy dummies rate are trending. It is impossible to argue that the unemployment rate is correlated with the policy dummies. The first two restrictive dummies were introduced at a time when the unemployment rate was low by historical standards. The liberalisation in 1981 also took place at a time of low unemployment although slightly higher than in 1974 and 1977. The dummies that capture Norway's relationship to the EU are also not related to contemporaneous variables in the model. The 1994 enlargement affected several countries and Norway's decision not to join might pos- sibly be related to economic factors (as in 1972 when there also was a no majority in a referendum).

The enlargements in 2004 and 2007 clearly had nothing to do with Norwegian politics.

The main empirical results are reported in Table 2.¹⁹ In the left part of this table we consider the unrestricted case and in the right part a restricted case. The restricted case is mainly obtained by excluding insignificant variables from the econometric specification.²⁰ However, when economic variables enter with a correct sign, we have retained them in the models even if the attached estimated slope parameters are not very significant. The restricted specification cannot be rejected when tested against the unrestricted specification using an LR-test.²¹ Hence, in the following we only comment on the restricted case.

As seen from Table 2 we obtain correct signs of the estimated effects of the lagged endogenous variable and the incentive variables. The effect of the lagged endogenous variable is large and highly significant. The lagged stock of immigrants from a specific country relative to the Norwegian population (log-transformed) enters significantly in the specification and with a positive sign as expected. GDP per capita relative to the level in Norway (with a two year lag and log-transformed) is included as suggested by theory, but its estimated slope coefficient only has a t-value of about 1.4 (in absolute value). The Norwegian unemployment rate enters significantly. An increase in the Norwegian unemployment rate decreases, ceteris paribus, immigration to Norway.

19We do not report estimates of the country-specific fixed effects and the country-specific linear trend effects in Table 2.

20 All the country-specific fixed effects have been retained, as well as country-specific trend variables with estimates with t- values higher than unity in absolute value.

21 The unrestricted model contains 319 unknown parameters including the variance of the error term and has a log-likelihood value equal to −3,493.99. The corresponding figures for the restricted model are 233 and −3,506.17. Thus using an LR-test statistic the restricted model cannot be rejected against the unrestricted model.

Table 2. Empirical analysis of immigration to Norway from the entire world. Unrestricted and restricted specification^a

Unrestricted case Restricted case Variable

Estimate t-value Estimate t-value

log(M/P)t-1 0.592 46.940 0.598 49.169

log(IS/PNOR)t-1 0.030 1.590 0.030 2.315

log(GDPCAP/GDPCAPNOR)t-2 -0.038 -0.899 -0.046 -1.413

URNORt-1 -0.060 -6.750 -0.059 -7.071

DUMCHILE 1.466 3.482 1.414 3.407

DUMLIBERIA 2.399 2.701 2.395 2.720

DUMSOMALIA 1.790 2.995 1.814 3.112

DNNORDIC×DDUM1974 -0.105 -3.479 -0.104 -3.694

DNNORDIC×DUM1977 -0.063 -2.186 -0.059 -2.148

DNNORDIC×DUM1981 0.073 2.662 0.075 3.354

DNNORDIC×DUM1991 -0.099 -3.232 -0.096 -3.593

DEEA×DUM1994 0.109 1.603 0.142 2.606

DREFUGEE×DUM1997 0.788 3.637 0.800 4.029

DNNORDIC×DUM1998 0.047 1.277 0.052 1.567

DNNORDIC×DUM1999 -0.190 -4.844 -0.189 -4.900

DUMMYIRAQ×DUM2000B -0.141 -0.408

DNNORDIC×DUM2000A -0.075 -1.952 -0.074 -1.953

DSCHENGEN×DUM2001A 0.123 1.727 0.140 2.059

DNNORDIC×(1-DSCHENGEN)×DUM2001B 0.143 4.572 0.144 4.823

DREFUGEE×DDUM2003 -0.088 -0.415

DEXTEU×DDUM2004 0.885 4.560 0.900 6.271

DVIETNAM×DUM2005 0.147 1.225 0.146 1.243

DVISA×DUM2006 0.009 0.283

DBULROM×DUM2007EU 0.493 2.060 0.502 2.129

DEXTEU×DUM2007A 0.080 0.347

DLIB×DUM2007B 0.062 1.580 0.073 2.487

DSTRICT×DUM2008 -0.190 -5.984 -0.189 -6.145

DTRANS×DUM2009A 0.254 1.058

DSWI×DUM2009B 0.132 0.227

Number of observations 4,220 4,220 R² 0.947 0.946

aLeft hand side variable log(M/P)t. For the definition of the variables in the text column see Table B1.

We find that the majority of the policy intervention variables enter with the correct sign. For some of the intervention variables we find no significant effects. In Table 3 we give a qualitative overview of the obtained results. For the immigration regulations introduced in 1974 and 1977, respectively, the correct negative sign is obtained. The liberalisation introduced in 1981 has as expected a positive effect. For the liberalization policy launched in 1991 we obtain a significant estimate with the wrong sign. As explained in Appendix C the policy changes in 1991 consisted of both restrictive measures and liberalisations so the total estimated effect is perhaps not surprising.

Table 3. Expected and estimated sign of coefficients for policy variables. Restricted specification^a

Variable Expected sign Estimated sign

DNNORDIC× ^DDUM1974 Negative Negative

DNNORDIC× ^DUM1977 Negative Negative

DNNORDIC×^DUM1981 Positive Positive

DNNORDIC×^DUM1991 Positive Negative

DEEA×^DUM1994 Positive Positive

DREFUGEE×^DUM1997 Positive Positive

DNNORDIC×^DUM1998 Positive Positive

DNNORDIC×^DUM1999 Positive Negative

DNNORDIC×^DUM2000A Positive Negative

DSCHENGEN×^DUM2001A Positive Positive

DNNORDIC×(1-DSHENGEN)×^DUM2001B Negative Positive

DEXTEU×^DDUM2004 Positive Positive

DVIETNAM×^DUM2005 Positive Positive

DBULROM×^DUM2007EU Positive Positive

DLIB×^DUM2007B Positive Positive

DSTRICT×^DUM2008 Negative Negative

aLeft hand side variable log(M/P)t. For the definition of the variables in the text column see Table B1.

A liberalization aimed at refugees was introduced in 1997. A correct sign is obtained for the estimated coefficient attached to this variable, and the estimate is significant. Also for the liberalization launched in 1998 we obtain the correct sign and a significant estimate. A wrong sign is obtained in connection with the liberalization in 1999. The Schengen area convention introduced in 2001 is expected to increase immigration to Norway from countries in the Schengen area but to lead to less immigration from the countries outside the Schengen area. Let us first consider the Schengen area. For this area we obtain the right positive sign, but, against intuition, the estimate of the effect for the non Schengen area is also positive and significant. The absolute value is equal to the estimate for the Schengen area.

In 2003 a stricter regime for family-reunion was introduced. This intervention is restricted to influence potential immigrants from Afghanistan, Iraq, Somalia and countries in former Yugoslavia. We are unable to find any negative effect of this intervention variable. In 2004 there was an extension of the EU/EEA area with some new East-European countries. The consequence was that people from these new countries obtained easier access to Norway. Hence, the sign of the estimated effect is in

accordance with our a priori expectation. The dummy that captures the positive immigration effect from Philippines and Iraq to Norway enters with the correct sign, but the effect is not very significant.

In 2007 there was another extension of the EU/EEA area since Bulgaria and Romania were included.

In accordance with our expectations we obtain a positive effect of this extension. The stricter demands for family reunion introduced in 2008 had, as expected, a significant negative influence.

We have also included dummies for Chile, Liberia and Somalia. A look at preliminary estimation results revealed that the residuals for these three countries were especially large in some years. Hence the dummy variables DUMCHILE, DUMLIBERIA and DUMSOMALIA are included to account for these large residuals.²² The estimates of the three attached parameters are all positive and significant.

Table 4. Empirical analysis of immigration to Norway from the entire world. Restricted specifi- cation. Model without incentive variables and model without trend variables^a

Variable

Reference (restricted)

Without incentive variables

Without trend variables

Estimate t-value Estimate t-value Estimate t-value log(M/P)t-1 0.598 49.169 0.605 51.447 0.755 72.798

log(IS/PNOR)t-1 0.030 2.315 0.057 5.264

log(GDPCAP/GDPCAPNOR)t-2 -0.046 -1.413 0.264 13.498

URNORt-1 -0.059 -7.071 -0.061 -7.005

DUMCHILE 1.414 3.407 1.466 3.511 1.371 3.126

DUMLIBERIA 2.395 2.720 2.318 2.616 2.206 2.456 DUMSOMALIA 1.814 3.112 1.864 3.178 0.882 1.686 DNNORDIC×DDUM1974 -0.104 -3.694 -0.088 -3.121 -0.082 -2.786 DNNORDIC×DUM1977 -0.059 -2.148 -0.052 -1.904 0.024 0.845 DNNORDIC×DUM1981 0.075 3.354 0.035 1.727 0.134 5.852 DNNORDIC×DUM1991 -0.096 -3.593 -0.222 -12.923 0.015 0.535

DEEADUM1994 0.142 2.606 0.138 2.520 0.082 1.708

DREFUGEE×DUM1997 0.800 4.029 0.848 4.249 0.202 1.673 DNNORDIC×DUM1998 0.052 1.567 0.161 5.482 0.106 3.086 DNNORDIC×DUM1999 -0.189 -4.900 -0.141 -3.675 -0.192 -4.705 DNNORDIC×DUM2000A -0.074 -1.953 -0.073 -1.918 -0.069 -1.735 DSCHENGEN×DUM2001A 0.140 2.059 0.136 1.999 0.249 3.952 DNNORDIC×(1-DSCHENGEN)×

DUM2001B

0.144 4.823 0.102 3.505 0.189 6.032

DEXTEU×DDUM2004 0.900 6.271 0.933 6.474 0.714 6.080 DVIETNAM×DUM2005 0.146 1.243 0.121 1.022 0.156 1.669 DBULROM×DUM2007EU 0.502 2.129 0.555 2.340 0.916 4.264 DLIB×DUM2007B 0.073 2.487 0.119 4.134 0.098 3.199 DSTRICT×DUM2008 -0.189 -6.145 -0.149 -4.899 -0.200 -6.164 Number of observations 4,220 4,220 4,220

R² 0.946 0.945 0.936

a Left hand side variable log(M/P)t. For the definition of the variables in the text column see Table B1.

In Table 4 we report estimation results for two special cases of the reference model. In the third column we report the estimates of a model where the parameters attached to the incentive variables are

constrained to zero. The main impression is that the parameter estimates attached to the policy intervention variables are not much changed qualitatively by the zero restrictions. The sign of the

22The binary variable DUMCHILE is one in 1987 and 1988 and zero in all other years and affects only Chile. The binary variable DUMLIBERIA is one in 2003 and 2004 and zero in all other years and affects only Liberia. The binary variable DUMSOMALIA is one in the years 1988−2010 and zero in all the years before 1988 and affects only Somalia.

estimates are the same as in the reference specification. So the estimates of the effects of the intervention variables seem to be fairly robust with respect to whether the incentive variables are included or not.

In the column next to the last of Table 4 we report the estimates of a model where all the country specific trend variables have been omitted. For this case we obtain a higher estimate of the coefficient attached to the lagged endogenous variable and a positive significant effect of the relative GDP-variable. Thus, the presence of country specific linear trends seems to be necessary in order to get the right sign of the relative GDP-effect. The model with omitted country specific linear trend variables contains 170 parameters (includeing the variance of the error term) and has a log-likelihood value equal to −3774.52. Thus if one tests this specification against the reference specification using an LR-test one obtains a χ2 value of 561.06. The associated degree of freedom is 65. Hence, the specification without country specific trends is clearly rejected.

Our main estimation method is weighted least squares with population as weights. The reason for this is that we are pooling countries that differ substantially in population size. We have also estimated the reference model with ordinary least squares. The results are reported in Table A1. Even if most of the estimates retain their sign they differ somewhat from those obtained when using weighted least squares with population weights and so does the estimation uncertainty. For instance the variable representing the immigration restrictions launched in 1977 still have the right sign, but the magnitude of the estimated slope coefficient of this variables has been almost halved and it has now turned insignificant. Thus, it makes a difference which estimation method that is used.²³

In the second column of Table B2 the effective number of observations for each country involved in the estimation of the main model is reported. For some of the countries the number of effective observations is rather low. In light of a potential problem of biased estimation stemming from few observations in the time dimension in dynamic models with fixed effects, cf. Nickell (1981), we have reestimated the main model after excluding countries with fever than 15 observations. The estimates in this case are reported in Table A2 and they show there is no substantial change in any of the estimates, which may imply that there is no “Nickell-bias”.

23 We have also estimated the reference model with weighted least squares using immigration weights. However, some of the results appear rather strange. The estimate of the slope parameter attached to the immigration stock now turns negative and besides the coefficient of the lagged endogenous variable is substantially lower than when weighted least squares is based on population weights. Finally, we have carried out weighted least squares using log population as weights. This variant produced results that resemble those obtained using ordinary least squares.

Changes in the income distribution

As commented on earlier in the paper changes in the income distribution in both the origin and destination country may influence immigration. It is relevant to ask whether this effect is important from an empirical point of view. For 101 of the 145 countries considered when estimating the reference model we have access to time series for the Gini-coefficient. Using this subsample we reestimated the reference model after having added a second order polynomial in the ratio between the Gini-coefficient of the origin country and Norway. Whereas the reference model was estimated using 4,220 observations the augmented model is estimated with 3,083 observations. The results are shown in Table 5. We obtain a significant positive estimate of the first order variable and a significant negative estimate of the quadratic term.²⁴ Within our sample an increase in the inequality of the income distribution of the origin leads, mainly, to, an increase in the immigration to Norway. In this augmented model the estimate of the income ratio is smaller and substantially less significant than in the reference model, the t-value now being only around 0.4 in absolute value. Generally the estimates of the common parameters in the augmented and reference model are similar. We now get a smaller and insignificant estimate of the coefficient attached to the intervention directed towards refugees from 1997. This is not surprising since some of the countries influenced by this variable are omitted from the subset of data used in the conjunction with the subsample estimation. However, by and large, including variables on income inequality for a smaller set of countries with appropriate data, does not change our conclusions with regard to the qualitative effects of policy interventions.

24 From Eq. (7) we have that GINIRATIOit denotes the ratio between the Gini-coefficient in country i and Norway in year t. In the estimated regression the effect of the variable is specified as ˆ₆ ˆ₇₂ ²

it it

GINIRATIO GINIRATIO

β +β ^{, where} ˆ6 0.935

β ^{= and} βˆ7^{= −}0.203. Note that the derivative is given by βˆ₆⁺2βˆ₇GINIRATIO. In our sample GINIRATIO varies between 0.697 and 2.915. An increase in GINIRATIO yields an increase in the immigration for most of the countries.

However, when evaluating the term βˆ₆⁺2βˆ₇GINIRATIO in the observed data points, we find that it is negative in at least one year for 24 countries.

Table 5. Empirical analysis of immigration to Norway. Models without and with time series of Gini-variables^a

Without Gini-variables With Gini-variables Variable

Estimate t-value Estimate t-value

log(M/P)t-1 0.598 49.169 0.583 40.391

log(IS/PNOR)t-1 0.030 2.315 0.034 2.220

log(GDPCAP/GDPCAPNOR)t-2 -0.046 -1.413 -0.017 -0.412

URNORt-1 -0.059 -7.071 -0.053 -5.729

GINIRATIO 0.935 4.753

GINIRATIO SQUARED -0.203 -3.601

DUMCHILE 1.414 3.407 1.478 3.320

DUMLIBERIA 2.395 2.720

DUMSOMALIA 1.814 3.112

DNNORDIC×DDUM1974 -0.104 -3.694 -0.092 -3.015

DNNORDIC×DUM1977 -0.059 -2.148 -0.054 -1.809

DNNORDIC×DUM1981 0.075 3.354 0.057 2.287

DNNORDIC×DUM1991 -0.096 -3.593 -0.106 -3.554

DEEA×DUM1994 0.142 2.606 0.147 2.488

DREFUGEE×DUM1997 0.800 4.029 0.516 0.745

DNNORDIC×DUM1998 0.052 1.567 0.069 1.896

DNNORDIC×DUM1999 -0.189 -4.900 -0.209 -4.889

DNNORDIC×DUM2000A -0.074 -1.953 -0.038 -0.902

DSCHENGEN×DUM2001A 0.140 2.059 0.150 2.030

DNNORDIC×(1-DSCHENGEN)× DUM2001B 0.144 4.823 0.153 4.592

DEXTEU×DDUM2004 0.900 6.271 0.933 5.832

DVIETNAM×DUM2005 0.146 1.243 0.168 1.178

DBULROM×DUM2007EU 0.502 2.129 0.476 1.723

DLIB×DUM2007B 0.073 2.487 0.061 1.846

DSTRICT×DUM2008 -0.189 -6.145 -0.205 -5.560

Number of observations 4,220 3,083 R² 0.946 0.952

a Left hand side variable log(M/P)t. For the definition of the variables in the text column see Table B1. Note that some of the variables in the text column have to be redefined when one considers estimation using data only for countries for which we have access to time series of Gini coefficients. For example DREFUGEE degenerates to an indicator dummy for Croatia, Macedonia and Slovakia since we, cf. Table B2, do not have income distribution data for Afghanistan, Iraq, Somalia, Bosnia Herzegovina, and Montenegro and Serbia.

The importance of the unemployment rate in the origin country

In the reference model the Norwegian unemployment rate enters as a significant explanatory variable with a negative sign. An interesting question is whether the unemployment rate in the origin country also plays a role. Unfortunately, we only have unemployment rates for a small group of selected countries, mostly OECD-countries. In Table 6 we consider a subsample estimation using data for 31 countries in which we add the foreign unemployment rate lagged one year as an additional regressor.²⁵ As is seen from the left hand part of Table 6 we obtain a significant positive estimate of the

unemployment level in the origin country and as before a negative coefficient for the Norwegian unemployment level. The difference in absolute value suggests that the two unemployment variables

25The countries are listed in Table B3.

should be specified as two separate variables in the regression. Just using the difference in the unemployment rates does not seem to be empirically valid. Note that for this subsample we obtain a significant negative estimate of the relative income variable as expected. However, we still struggle with some of the signs of the effects of the intervention variables, for instance the effects related to the two liberalization interventions in 1999 and 2000. In the right hand part of Table 6 we report results from a specification in which we also have added a second order polynomial in the Gini-coefficient ratio but the two involved variables are highly insignificant for this subset of countries.

Table 6. Empirical analysis of immigration from countries for which one observes the origin unemployment level^a

Without Gini-variables With Gini-variables Variable

Estimate t-value Estimate t-value

log(M/P)t-1 0.602 23.194 0.599 22.243

log(IS/PNOR)t-1 0.057 1.741 0.071 1.940

log(GDPCAP/GDPCAPNOR)t-2 -0.182 -2.656 -0.221 -2.853

URNORt-1 -0.041 -4.571 -0.037 -3.907

URt-1 0.009 2.505 0.008 2.153

GINIRATIO 0.002 0.005

GINIRATIO SQUARED -0.007 -0.065

DNNORDIC×DDUM1974 -0.056 -1.719 -0.060 -1.786

DNNORDIC×DUM1977 -0.070 -2.436 -0.073 -2.509

DNNORDIC×DUM1981 -0.042 -1.686 -0.050 -1.916

DNNORDIC×DUM1991 0.093 3.184 0.084 2.786

DEEA×DUM1994 0.058 1.945 0.061 1.985

DNNORDIC×DUM1998 -0.045 -1.245 -0.039 -1.057

DNNORDIC×DUM1999 -0.108 -2.481 -0.106 -2.389

DNNORDIC×DUM2000A -0.116 -2.713 -0.118 -2.684

DSCHENGEN×DUM2001A 0.202 4.410 0.191 4.001

DNNORDIC×(1-DSCHENGEN)× DUM2001B 0.060 1.667 0.048 1.305

DEXTEU×DDUM2004 0.815 6.877 0.794 6.455

DLIB×DUM2007B -0.016 -0.408 -0.015 -0.374

DSTRICT×DUM2008 -0.063 -1.479 -0.016 -0.338

Number of observations 1,052 982

R² 0.983 0.979

a Left hand side variable log(M/P)_t. For the countries included in this estimation see Table B4. For the definition of the variables in the text column see Table B1.

Note that some of the variables in the text column are redefined when one considers estimation using data only for countries for which one has access to origin unemployment rates. For example DEXTEU is now one for Estonia, Hungary, Poland and Slovakia and zero for all other countries included in the estimation of the econometric relation.

Some counterfactual exercises

We now use the estimated model for counterfactual analysis. We will consider two policy changes.

Simulation I tries to answer how immigration to Norway would have changed if the 1973-policy regime had been maintained in the subsequent years. The second policy analysis (Simulation II) addresses how Norwegian membership in the European Economic Area (EEA) and the Schengen area has affected immigration. These analyses are not without problems. First, we are unable to obtain the

“correct” sign of all the estimated parameters related to the intervention dummies, cf. Table 3.

Second, we implicitly will have to assume that the estimates of the slope parameters are not affected by the counterfactual situation. Third, the dataset is, as mentioned earlier, unbalanced, which creates problem for the dynamic simulation of all the countries in the model. Finally, it is a very partial exercise in that all other variables of the model are assumed unaffected. If immigration is higher, several of the right hand side variables might possibly be affected too and these changes are not included in the simulations. One obvious example is that a change which increases immigration will most likely also increase the stock of immigrants unless there is a similar increase in emigration. The latter effect is ignored in these simulations. The countries included in, respectively, Simulation I and Simulation II are listed in the two last columns of Table B2. The point of departure is the reference model.

In Simulation I we study the “global” intervention effects and start the dynamic simulations in 1974. As a reference we simulate a model that corresponds to the restricted case in Table 2. For each year we deduce the total number of immigrants from the 70 countries indicated in the column next to the last in Table B2. In the counterfactual simulation we set, cf. Table B1, the following policy variables to zero:

DDUM1974, DUM1977, DUM1981, DUM1997, DUM1998, DUM2007B and DUM2008.²⁶ The simulation results are reported in Table 7. In the first column we report the reference path, whereas the counterfactual path is reported in the second column. The two last columns contain the difference in immigration between the counterfactual and the reference path in absolute and relative terms. In the 1970s we note the impact of the restrictions launched in 1974 and 1977. Our estimate is that immigration to Norway due to these two policies was reduced by 28 percent by 1980. The effects increase over time due to the lagged responses of the policies. The 1981-liberalisation reduced the effects of the more restrictive policies significantly during the first half of the 1980s. The policies of the 1970s and early 1980s thus seem to have reduced total immigration by roughly 16 percent. The liberalisation in 1998 increases immigration further and by early 2000s the total effects of the mentioned policies have reduced immigration by only 5 percent compared to policies that were in place before 1974.

If we look at the accumulated changes over three decades the total effect on immigration has been considerable. By 2010 total immigration was reduced by nearly 116 000 persons due to these

immigration policies according to our model. But we should note the partial character of the simulations.

Although some of the immigrants would have emigrated again, the stock of immigrants has been

26 Since the variables DNNORDIC×DUM1991, DNNORDIC×DUM1999, DNNORDIC×DUM200A and

DNNORDIC×(1-DSCHENGEN)×DUM2001B enter with the wrong sign, we include them both in the reference and counterfactual simulations. Thus we refrain from interpreting these variables as intervention variables.

negatively affected by the policies. A lower stock of immigrants would have reduced the number of immigrants further since stocks of immigrants in Norway reduce transaction costs. On the other hand a higher number of immigrants may have increased unemployment and led later to fewer immigrants.

These arguments simply add up to the need of analysing the issue within a much more complete “model”

of the Norwegian society.

Table 7. Counterfactual analysis of immigration to Norway (70 countries). Simulation I^a Year Reference

pat h

Counterfactual path

Absolute differ

ence

Difference in perce

nt

1974 16,969 18,066 1,097 6.5

1975 17,626 19,584 1,958 11.1

1976 17,807 20,301 2,494 14.0

1977 17,613 21,167 3,554 20.2

1978 18,023 22,360 4,337 24.1

1979 18,165 22,978 4,813 26.5

1980 18,350 23,497 5,147 28.0

1981 19,953 24,589 4,636 23.2

1982 20,781 25,019 4,238 20.4

1983 20,830 24,730 3,900 18.7

1984 20,147 23,727 3,580 17.8

1985 20,229 23,716 3,487 17.2

1986 21,296 24,907 3,611 17.0

1987 23,660 27,686 4,026 17.0

1988 25,999 30,492 4,493 17.3

1989 24,482 28,636 4,154 17.0

1990 21,929 25,639 3,710 16.9

1991 19,220 22,380 3,160 16.4

1992 17,739 20,613 2,874 16.2

1993 16,776 19,476 2,700 16.1

1994 16,966 19,732 2,766 16.3

1995 17,886 20,831 2,945 16.5

1996 19,298 22,502 3,204 16.6

1997 21,889 25,392 3,503 16.0

1998 25,959 28,992 3,033 11.7

1999 26,463 28,773 2,310 8.7

2000 25,975 27,752 1,777 6.8

2001 28,362 30,083 1,721 6.1

2002 30,417 32,102 1,685 5.5

2003 31,635 33,276 1,641 5.2

2004 32,676 34,344 1,668 5.1

2005 34,633 36,409 1,776 5,1

2006 37,572 39,152 1,580 4,2

2007 47,901 47,908 7 0,0

2008 52,562 55,407 2,845 5,4

2009 55,288 60,296 5,008 9,1

2010 57,657 63,934 6,277 10,9

Sum 960,733 1,076,448 115,715 12,0

aThe reference path corresponds to dynamic simulation starting in 1974 using the estimated parameters reported for the restricted case in Table 2. For the counterfactual path we set the values of the intervention variables DDUM1974, DUM1977, DUM1981, DUM1997, DUM1998, DUM2007B and DUM2008 to zero in all years and perform dynamic simulation. The countries involved are listed in the column next to the last in Table B2.